A tutor that talks back, and never gets the answer wrong
Open Super Teacher and a small animated face starts talking to your six-year-old by name. It asks a question. The child answers out loud - "absolutely," maybe, instead of a clean "yes" - and the app understands, then bends the next moment toward enrichment or back toward the basics. It feels like magic. It is, deliberately, not.
Tim Novikoff runs the company that makes it, and he will tell you the part most edtech founders bury in the FAQ: there is no large language model improvising under the hood. "Everything that Super Teacher says and shows to children was crafted in advance by a real, human teacher," the company says. The app leans on three older, sturdier technologies - speech transcription to hear the child, natural language processing to read intent, and a synthetic voice to narrate - and it routes between lessons a teacher wrote by hand. No ChatGPT. No hallucinated facts told to a kid who can't yet tell the difference.
That choice is the whole personality of the thing. For elementary learners, Novikoff has decided, "always correct" beats "endlessly flexible." The system is deterministic on purpose. A startup in 2026 bragging about not using a language model is a contrarian bet, and it is the bet he made.
It's by far the most effective intervention that can be provided to kids for education.- Tim Novikoff, on one-to-one tutoring
The argument behind the product is older than the product. Decades of research keep landing on the same finding: a good tutor, working one-to-one, is about the most powerful lever in education. The catch has always been arithmetic. A human tutor at $50 an hour is a luxury good. Novikoff's pitch is that software can carry the structure of tutoring - the listening, the branching, the patient re-explaining - down to roughly the price of a couple of coffees. Super Teacher costs $15 a month, or $10 on an annual plan. The families signing up are not, mostly, the families who could already afford a private tutor.
Acting school, then a math degree, then a cartoon
Before any of this, there was a microphone and a kid. In the mid-1990s Novikoff voiced Jeffy, a minor character on MTV's cult animated series Daria. He grew up a child actor, and when he got to New York University he did the sensible thing and majored in acting - then, less sensibly, added a second major in mathematics. Acting and algorithms, side by side on the same transcript. He has been bridging those two ever since, and he says so plainly: the reason entrepreneurship fit him is that "it's both technical and creative."
The math eventually won a seat at the table. After NYU he taught - four years of high-school mathematics at Stuyvesant, one of the most competitive public schools in the country, by way of the NYC Teaching Fellows program and a master's in education. Standing in front of teenagers all day is where the obsession started. How often should a student review something before it slips away? Too soon and you waste the effort. Too late and it's gone.
Entrepreneurship was a better fit for me because it's both technical and creative.- Tim Novikoff
He turned the question into a smartphone app called Flash of Genius, vocabulary flashcards for the SAT, and then he turned the app into a PhD. At Cornell he studied algorithmic education theory under two heavyweights - the mathematician Steven Strogatz and the computer scientist Jon Kleinberg. In 2012 the three of them published "Education of a Model Student" in the Proceedings of the National Academy of Sciences, a paper that does something audacious: it treats a student's memory as a set of constraints and asks how fast a teacher can pour in new material without the old material leaking out. The answer comes with named strategies - the Recap Method, the Slow Flashcard Method, Hold-Build - and a sobering footnote: some learners need so much review that, in the idealized model, you can never introduce anything new at all.
A Queen cover band, a Halloween party, and an exit to Google
The straight line from flashcards to AI tutor has a loop in it, and the loop is shaped like a guitar. At his own Halloween party, Novikoff wanted to film a Queen cover band from several angles at once and stitch the footage together. There was no good app for that. So at a three-day Cornell startup event in 2011 he started building one.
That became Fly Labs. The flagship app, Fly, shipped in the summer of 2014, crossed a million downloads, and landed on best-of-2014 lists from both Apple and The Boston Globe. The company put out Crop, which rescued people from accidental vertical video, and Clips, for splicing footage together. It won The Next Web's pitch competition and made Business Insider's list of New York's 25 hottest startups - a team of five, building "apps for everybody," democratizing video editing for people who would never open professional software.
In 2015, Google bought it. Novikoff stayed for years, working on the Memories and editing features inside Google Photos and then running Colab, the notebook tool that a generation of data scientists and machine-learning engineers now treat as a default. He had a comfortable, important job at one of the most powerful companies on earth. He left it to go back to teaching six-year-olds.
The Fly Labs origin story is a cover band. The idea was multi-angle concert footage from a phone. Google ended up owning the result.
His PhD advisers were Strogatz and Kleinberg - the kind of committee most applied mathematicians only cite.
For over a decade he taught iPhone app development at Cornell Tech, on both the NYC and Ithaca campuses, while running companies.
The PNAS paper named its strategies like wrestling moves: Recap, Slow Flashcard, Hold-Build.
The same job he always had
Read his career sideways and there is no detour at all. The acting taught him performance and voice. The teaching gave him the problem. The math gave him the model. The video startup taught him how to ship a delightful consumer app and sell it. Google taught him AI at scale. Super Teacher is all of it pointed at one question he has been chewing on since Stuyvesant: how do you give every kid the thing that actually works?
Co-founded in 2022 in New York with Krzysztof Kulewski, the company is staffed with the people you'd want building this - classroom teachers who have run preschool through second grade, mathematicians, and Google-trained engineers, many of them parents of young children themselves. In October 2025, Super Teacher demoed onstage at TechCrunch Disrupt in San Francisco as a Startup Battlefield Top 20 finalist. Roughly 20,000 families had signed up. Public schools in New York, New Jersey, and Hawaii were using it. The plan from here is unglamorous and clear: more grades, younger and older, and more districts.
Novikoff is careful about what he's claiming. He frames the AI tutor as a tool for teachers, not a replacement for them - which, from someone who spent four years at the front of a classroom and has watched the hype cycle from inside Google, reads less like marketing and more like a line he actually believes.
Everything that Super Teacher says and shows to children was crafted in advance by a real, human teacher.- Super Teacher's design principle
There is a tidy irony in a former Google AI lead building the AI product that pointedly won't let the AI improvise. But it tracks with the man who double-majored in acting and math, who once measured the half-life of a memorized word, who keeps choosing the harder, more honest version of the thing. The cartoon voice grew up to write the script himself.
Things that don't fit the resume
- He voiced Jeffy on MTV's Daria as a kid.
- Double major at NYU: acting and mathematics.
- A Halloween-party cover band sparked the app Google bought.
- His company refuses to use ChatGPT-style models on purpose.
- The tutors greet each child by name with a synthetic voice.
- His flashcard side project turned into a peer-reviewed PNAS paper.